import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler, MinMaxScaler, RobustScaler, MaxAbsScaler, PowerTransformer
# Define the features
features = ['danceability', 'energy', 'loudness', 'speechiness', 'acousticness', 'instrumentalness', 'liveness', 'valence', 'tempo']
# Extract features from the dataframe (assuming spotify5k_df is defined elsewhere)
# X = spotify5k_df[features]
# Number of clusters
k = 4
# List of scaler names for plotting and labeling
scaler_names = ['Raw', 'StandardScaler', 'MinMaxScaler', 'RobustScaler', 'MaxAbsScaler', 'PowerTransformer']
# Dictionary to store Within-Cluster-Sum-of-Squares (WCSS) for each scaler
wcss_dict = {}
# Define colors for better readability
colors = sns.color_palette("tab10")
# Dictionary to store cluster centers for each scaler
cluster_centers = {}
# Iterate over each scaler
for i, scaler in enumerate([None, StandardScaler(), MinMaxScaler(), RobustScaler(), MaxAbsScaler(), PowerTransformer()]):
# Scale the features if scaler is not None
if scaler is not None:
X_scaled = scaler.fit_transform(X)
scaler_name = scaler_names[i]
else:
X_scaled = X
scaler_name = 'Raw Data'
# Initialize KMeans clustering
kmeans = KMeans(n_clusters=k, random_state=42, n_init=10)
kmeans.fit(X_scaled)
# Calculate WCSS
wcss = kmeans.inertia_
wcss_dict[scaler_name] = wcss
# Store cluster centers
cluster_centers[scaler_name] = kmeans.cluster_centers_
# Plot radar chart for cluster centers
plt.figure(figsize=(10, 6))
angles = np.linspace(0, 2 * np.pi, len(features), endpoint=False).tolist()
angles += angles[:1] # Ensure closed loop
ax = plt.subplot(111, polar=True)
# Plot each cluster's centroid
for idx, center in enumerate(kmeans.cluster_centers_):
values = np.round(np.concatenate((center, [center[0]])), 2) # Round to two decimal points
ax.plot(angles, values, marker='o', linestyle='-', color=colors[idx], linewidth=2, label=f'Cluster {idx+1}')
# Fill the area enclosed by each cluster's centroid with a light blue shade
ax.fill(angles, values, color=colors[idx], alpha=0.25)
# Set the labels for each axis
ax.set_yticklabels([])
ax.set_xticks(angles[:-1])
ax.set_xticklabels(features, fontsize=10)
plt.title(f'Radar Chart for Cluster Centers ({scaler_name})', loc='left', fontsize=12, pad=20)
plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.1), ncol=k)
plt.show()
# Create DataFrame to display WCSS for the current scaler
scaler_df = pd.DataFrame(list(wcss_dict.items()), columns=['Scaler', 'WCSS'])
# Display the table for the current scaler
print(f"{scaler_name} chart")
print(f"Table for {scaler_name}:")
display(scaler_df)
# Find the best scaler based on the minimum WCSS
best_scaler = min(wcss_dict, key=wcss_dict.get)
print(f"\nBased on the Within-Cluster-Sum-of-Squares (WCSS), the best scaler to use is {best_scaler}. Using this scaler helps in minimizing the WCSS, indicating better cluster formation.")
print(f"\nChoosing an appropriate scaler is crucial as it affects the clustering results. With {k} clusters, it is recommended to use the {best_scaler} scaler to create the clusters.")
# Add explanation points based on the WCSS score
print("\nAdditional points: \n1) A lower WCSS score indicates that the data points within each cluster are closer to their respective centroids, implying more compact and well-separated clusters. \n2) By selecting the scaler that yields the lowest WCSS, we aim to achieve the most meaningful and distinct cluster separation. \n3) The shaded area in each radar chart represents the coverage or extent of each cluster's features in the scaled feature space. It visually demonstrates how different clusters vary in their feature composition and distribution.")
